"my most controversial opinion is that I think that AI is as big a deal as the internet or mobile and only as big a deal as the internet or mobile" - Benedict Evans [00:02:52]
"imagine you're an accountant seeing the first software spreadsheets in the late '70s this is mind-blowing... but if you were a lawyer looking at that... you'd think well that's very clever and my accountant should see this but that's not what I do" - Benedict Evans [00:07:46]
Disclaimer: Orignal content owned by or sourced from third parties. It does not represent the views of 'Nuggets' platform or it's team. AI is used extensively across this platform including for summaries. Accuracy is not guaranteed, there can be mistakes. Any info or content on this platform is not a financial, legal, or investment advice. Do your own research. Refer for complete disclosures:- Terms of Use · Full Disclaimer
"if you make it cheaper to do something what happens do you do the same for less money or do you do more for the same amount of money or do you do more for more money because you've got new ROI" - Benedict Evans [00:13:49]
"what Amazon does is get you the SKU but knowing what SKU you want is another job" - Benedict Evans [00:15:16]
"you can't look at a senior partner at a law firm and say 'Well 17% of their work could be automated this is horshit.'" - Benedict Evans [00:00:32]
"every time we have a new technology... it automates away a bunch of jobs and then that automation... unlocks a bunch of new jobs and so you... don't know the new job because it doesn't exist yet" - Benedict Evans [00:18:58]
"the chatbot is a blank screen in a jagged edge like what am I supposed to do and what will work and that's a big problem and the solution to that problem is to wrap it in in use cases" - Benedict Evans [01:11:00]
Speakers & Credentials
Lenny Rachitsky: Host of Lenny's Podcast.
Benedict Evans: Independent Technology Analyst. Formerly a long-time partner and in-house analyst at Andreessen Horowitz (a16z). Veteran sell-side equity and telecoms analyst. Publisher of a widely-read technology newsletter and annual macro-trend presentations.
1. Executive Summary
The "1997 Internet" Baseline: AI is an unequivocally massive platform shift, but practically speaking, we are operating in an environment akin to 1997. Most use cases haven't been built, widespread corporate integration will take years, and predicting specific industry impacts remains speculative.
The Job Automation Fallacy: Discussions surrounding imminent, total job replacement fail to account for the Jevons Paradox (price elasticity). Making tasks cheaper historically leads to industries doing more of those tasks and developing entirely new workflows, rather than eliminating workforces.
The Task vs. Job Distinction: A foundational misunderstanding of AI's threat is conflating an isolated task (making a 75-slide PowerPoint deck) with a job (the strategic consulting, political navigation, and diagnostic analysis that necessitates the deck).
Commoditization & Value Capture: The foundational model layer risks rapid commoditization. Similar to how the telecom industry powers the mobile revolution without capturing its exponential profit growth, AI labs face the threat of becoming low-margin utilities while the value accrues at the application and distribution layers.
The Imperative to Engage: Despite justified concerns surrounding deepfakes, copyright, and market disruptions, retreating into anti-AI sentiment provides only a false sense of moral superiority. The only actionable path forward is aggressive experimentation and engagement with the tools to understand their specific utility.
2. Chronological Table of Contents
00:00:00 - Introduction & The Baseline Stance on AI's Impact
00:06:26 - The 1997 Internet Analogy & Spread of Adoption
00:09:46 - AI in Professional Services: Why Consultants are Thriving
00:12:32 - Task vs. Job & The Fallacy of Linear Automation
00:18:01 - The "Job Apocalypse" & Historical Precedents
00:26:06 - AGI, Definitions of AI, and Re-drawing the Boundaries
00:29:46 - TAM Expansion & The Utilities Paradox (Value Capture)
00:43:00 - The Supremacy of Distribution Over Commodity Models
The 1997 Internet Paradigm & Adoption Realities [00:02:52]
Historical Sizing: Evans insists AI is exactly as big of a deal as the internet or mobile technology [00:02:52]. This is meant to combat both skeptics who dismiss it and doomers who think it's an unprecedented apocalyptic event.
The "VisiCalc" Moment: Right now, certain demographics (like software engineers) are reacting to AI the way accountants reacted to VisiCalc (the first spreadsheet) in the late 1970s. For them, it instantly revolutionized their workflows [00:07:46]. Meanwhile, other professionals (like lawyers looking at a 1970s spreadsheet) recognize the tech is impressive but fail to see immediate personal utility.
Cost and Friction: In the 1970s, realizing the value of an Apple II, monitor, and printer cost roughly $10,000 to $15,000 adjusted for inflation [00:08:12]. Today, the cost barrier is effectively zero because AI leverages existing global hardware infrastructure, which exponentially accelerates consumer adoption.
The Generational Divide: A recent survey of 13-18 year olds reveals the jagged edge of adoption: only 15-20% are daily active users, another 20% are weekly active users, leaving a massive 60% of that demographic stating they simply aren't using AI tools [00:08:54].
Task vs. Job: The Economics of Automation [00:12:32]
The Elevator Operator Fallacy: Automation occasionally destroys whole roles (like manual elevator operators replaced by automated buttons in the 1950s) [00:13:14]. However, in most knowledge fields, AI simply compresses the task, it doesn't eliminate the job.
The Price Elasticity / Jevons Paradox: If a core task becomes vastly cheaper, demand for the output generally scales proportionately [00:13:49]. Evans notes that prior to Excel, junior investment bankers worked grueling hours; after Excel, they didn't start going home at lunchtime on Fridays—they just produced vastly more models for the same amount of money [00:14:08].
The E-Commerce (SKU) Analogy: Amazon is a machine designed to fetch a specific SKU, but realizing which SKU you need is an entirely separate cognitive process [00:15:16]. Claude Code can instantly generate a feature, but figuring out product-market fit requires a human.
Why AI Won't Kill McKinsey: AI grifters claim they can generate a 75-slide McKinsey deck using Claude. But executives do not hire Bain, BCG, or McKinsey for the PowerPoint itself. They hire them to interrogate political dynamics, navigate enterprise workflows, and talk directly to users to figure out why the business is failing in the first place [00:15:59].
Value Capture: Will Foundation Models Become Utilities? [00:32:40]
The Telecom Cautionary Tale: The global mobile telecom industry generates ~$1 Trillion annually and spends ~$200 Billion (~20% of revenue) on raw CapEx [00:33:50]. Global mobile data consumption has scaled by 1,500 to 2,000 times since 2010, yet telecom stocks have been flat for 25 years because they are ultimately a low-margin commodity [00:34:06]. The massive value of the mobile revolution was captured by companies operating up the stack (Apple, Meta, Uber).
The Commodity Risk for OpenAI and Anthropic: Evans challenges Sam Altman’s vision of selling intelligence "on a meter" like electricity, noting utilities have notoriously poor margin structures. When consumers buy a Bosch washing machine, Bosch doesn't pay a cut to the electric company [00:33:02]. If foundational models lack strong network effects or radical differentiation, fierce competition will drive the API costs to marginal cost, squeezing margins [00:36:00].
Distribution as the Ultimate Moat: If underlying models are effectively rendering engines (like web browsers), then distribution outweighs raw performance [00:44:01]. Google's aggressive integration of Gemini into Android, and Meta's integration of Llama across their massive app ecosystem, proves that an "adequate" product with unparalleled distribution can often stall out technically superior challengers [00:45:03].
Debunking the Resource Drain Narrative: Widespread panics regarding AI's resource consumption lack context. A rigorous 2024 Livermore Lab study estimated US data center water consumption at roughly 0.017% of total US water usage [00:49:16]. Data centers currently account for ~5% of US energy usage, growing at roughly 1% a year [00:49:39].
The Black Box of Usage Data: A major driver of public anxiety is a sheer vacuum of transparency. Model labs refuse to release meaningful usage metrics or DAU (Daily Active User) data for flagship products like ChatGPT. Consequently, macro-economic impacts must be clumsily reverse-engineered by economists [00:50:47].
Historical Precedents of Tech Abuse: AI introduces novel vectors for harm (e.g., teenagers using AI to instantly generate deepfake revenge porn) [00:56:23]. However, technology ruining lives is not new. Evans cites the UK Post Office scandal, where a bug-ridden 1970s Fujitsu computer system falsely flagged shortfalls, leading to the wrongful imprisonment and bankruptcy of hundreds of franchise owners [00:57:14].
The Reference Vault
4. Data & Figures
Data Point
Value
Context
Timestamp
First-Gen PC Deployment Cost
$10,000 - $15,000
Adjusted cost of an Apple II, monitor, and printer required to run early spreadsheet software (VisiCalc).
The "VisiCalc" Adoption Model: Evaluates how disruptive tech initially scales. Software engineers reacting to AI today are accountants reacting to VisiCalc in 1978. Everyone else is a lawyer looking at VisiCalc—recognizing it as impressive, but failing to grasp its immediate utility to their own workflow. [00:07:46]
Task vs. Job (The SKUs Analogy): A mental model distinguishing between automation of execution versus automation of intent. Amazon excels at delivering a SKU, but cannot tell you which SKU you actually need. Claude can generate code/slides (the task), but cannot determine enterprise strategy or political viability (the job). [00:15:16]
The Jevons Paradox (Price Elasticity of Labor): If a technological advancement drastically lowers the cost of a resource (or task), society will respond by drastically increasing its consumption of that resource. Applied to AI, cheaper coding will not eliminate coders; it will lead to an exponential increase in software production. [00:13:43]
The Utilities Paradox (The Telecom Model): A framework predicting value capture. Just as global mobile carriers invest billions into raw infrastructure without capturing the exponential software value generated on top of it, AI model labs risk becoming undifferentiated, low-margin utilities powering high-margin applications. [00:33:50]
The "Uber vs. Airbnb" Disruption Test: A model for projecting secondary effects. Uber replaced taxis and massively expanded the market. Airbnb didn't destroy hotels; it carved out an adjacent parallel market because corporate travel fundamentally requires hotel infrastructure. AI's impact on industries will be highly non-linear and context-dependent. [01:05:03]
6. Anecdotes
The IBM "150 Extra Engineers" Ad (1950s): Evans highlights a 1950s print ad showing men in white shirts promoting a fridge-sized IBM "Electronic Calculator." The tagline promises "150 extra engineers." Evans notes this is identical to the exact value proposition pushed by current AI companies (like Claude Code), demonstrating the cyclical nature of tech marketing. [00:23:31]
The Barcode and the Food Marketing Institute: To build a chart on supermarket SKU expansion enabled by the barcode, Evans spent two hours on Google today. In 1994, building the same chart would have required days of long-distance calls to physical libraries and expensive reports, proving how deeply we have naturalized previous paradigm shifts. [00:24:28]
The 1999 Dot-Com Pitch: As a junior analyst in 1999, Evans listened to an elaborate dot-com pitch from an online computer parts seller. A senior banker bluntly shattered the hype, stating, "It's a low-margin reseller with a one-time sale." Evans applies this same ruthless logic to modern AI labs operating as undifferentiated infrastructure. [00:39:47]
The UK Post Office / Fujitsu Scandal: To illustrate that algorithmic harm is not unique to modern AI, Evans points to a disastrous 1970s POS system built by Fujitsu. Bugs in the code showed cash shortfalls, leading the UK Post Office to wrongfully prosecute, imprison, and bankrupt hundreds of franchise owners who were completely innocent. [00:57:14]
The Chicago Meatpackers: A historical example of logistics driving disruption. When it became cheaper to ship a live cow from New York to Chicago, slaughter it, pack it, and ship the meat back to New York via refrigerated railcars than to butcher it in NY, the entire market shifted. Evans compares this logistical inversion directly to modern broadband and computing dynamics. [01:13:08]
Pre-iPhone Mobile Hardware: Evans collects old phones to remind himself of technological convergence. He cites the Ericsson Shark fin phone (1998) and a color-screen Japanese iMode phone (2001) as examples of how industries wildly experiment with form factors before standardizing on a dominant, boring design (the glass rectangle). [01:17:10]
7. References & Recommendations
People
Marc Andreessen: Cited regarding software eating the world, TAM expansion, and the Uber vs Hotel disruption analogies. [00:29:46]
Sam Altman: CEO of OpenAI. Evans critiques his assertion that OpenAI will sell intelligence as an electrical utility. [00:32:40]
Dario Amodei: CEO of Anthropic. Mentioned in the context of taking macro-economic labor predictions from model builders with a grain of salt. [00:18:01]
Larry Tesler: AI Scientist quoted for the definition: "AI is whatever machines can't do yet." [00:27:23]
Steven Sinofsky: Former Windows executive at a16z. Noted for the maxim: "Incumbents always try and make the new thing a feature." [00:42:49]
Jonathan Swift: English satirist, quoted regarding the impossibility of reasoning someone out of a position they didn't reason themselves into, applied to AI culture wars. [00:52:36]
Eric Schmidt: Former Google CEO. Mentioned as receiving boos at a commencement speech for mentioning AI, highlighting growing anti-AI sentiment. [00:48:30]
Books & Literature
Three Men in a Boat by Jerome K. Jerome: A late-19th-century British comedy Evans cites as the source for his mental model of "I have an anecdote for everything." [01:12:40]
Nature's Metropolis by William Cronon: Highly recommended for understanding how physical logistics, packetization, and rail networks mirror the business dynamics of broadband and digital platforms. [01:12:54]
Media & Pop Culture
The Seventh Seal: The classic Ingmar Bergman film. Evans notes it as an example of media that seems impossibly intimidating/boring but is actually brilliant and worth watching over endless modern content. [01:14:13]
Pete Holmes: Stand-up comedian cited by Lenny regarding the irony of AI replacing art and creativity while humans still clean up streets. [01:10:25]
Companies, Products, & Institutions
McKinsey, Bain, BCG, Accenture, Infosys: Used as proxies for complex "Jobs" that cannot be instantly automated by an LLM generating a PowerPoint. They represent the "consulting" layer that model labs are surprisingly investing in. [00:15:59]
Fujitsu & The UK Post Office: Central entities in the major historical tech scandal used as an analogy for algorithmic bias and blind corporate faith in systems. [00:57:14]
Lawrence Livermore National Laboratory: Responsible for the late 2024 study tracking data center water consumption, debunking media panics. [00:49:16]
Food Marketing Institute: The trade group tracking the explosion of grocery SKUs driven by the invention of the barcode. [00:24:28]
O*NET: The US Government database that attempts to categorize job tasks. Evans views their attempts to mathematically score a job's "AI exposure" as fundamentally flawed logic. [01:02:39]
Apple (Cupertino) & WWDC: Mentioned frequently regarding Apple Intelligence's integration of Gemini and the power of on-device distribution over raw model supremacy. [00:46:24]
Ericsson & iMode: Early mobile phone manufacturers/ecosystems Evans collects to study the era before hardware form-factor convergence. [01:17:10]
8. The Bottomline (by AI)
The prevailing panic regarding imminent, total job replacement relies on a flawed, static view of economics that ignores the Jevons paradox; making tasks cheaper historically leads to an explosion of output and novel workflows, not a cessation of human labor. Furthermore, foundation model developers risk repeating the mistakes of the telecom industry—spending billions on capital expenditures only to become low-margin, commoditized utilities while the true financial upside is captured by the application and distribution layers built on top of them. For executives and builders, the strategic imperative is to stop obsessing over raw model intelligence and focus entirely on mastering distribution, deeply integrating AI into complex, multi-step business logic that generic chatbots cannot replicate.
"Brookfield's the largest infrastructure owner in the world... We drew a pipeline and we showed all the different components of the payments ecosystem on a pipeline and said it's like a pipe that moves any commodity except what it's moving…
Current Computing Footprint
~800M PCs / 5.5B - 6B Smartphones
Current global computing install base driving AI adoption speeds.